Expert Systems

1
Expert Systems AI

• MSC 636
• Winter 2003

2
What is Knowledge?

• Data
• Facts, measurements, or observations
• Specific accuracy
• Information
• Data organized to be useful relevant for a
  particular problem
• Personal relevance for problem
• Knowledge
• Interpretation of instincts, ideas, rules,
  relationships, procedures, information about a
  particular problem domain
• Evolves understanding of domain

3
Knowledge-based DSS

• Knowledge is everything you need to know to help
  solve groups of problems within a domain a
  mental construction more abstract dynamic than
  information
• Holsapple Whinston (1988)
• Declarative (descriptive) knowledge
• Procedural knowledge
• Reasoning knowledge
• Knowledge is implicit or explicit (or both)

4
Variable emphasis on knowledge, data, or models
5
Artificial Intelligence

• Artificial intelligence is supportive CS research
  domain
• Enable computers to simulate the reasoning
  processes performed by humans
• Intersection of computer science cognitive
  psychology
• Human reasoning processes
• Categorization
• Rules heuristics
• Past experiences
• Expectations

6
AI vs. Traditional Computing

• AI differs from conventional computing
• Reasoning (inferencing)
• Rule-based heuristic logical systems
• Pattern recognition
• Voice, fingerprints, faces, signatures
• Improved algorithms (Rete)
• Learning (self-evolving)
• Biological analogies, ANN, GA
• Classification categorization
• Handles incomplete data/linguistic ambiguities
• Fuzzy logic

7
Expert Systems

• Knowledge-driven DSS common in business
  applications
• Expert-derived rule-based knowledge base
• Database stores facts (temporarily stored in
  workspace/blackboard)
• Knowledge base stores relationships rules
• Inference engine processes facts rules directs
  processing (reasoning)
• Interface manages communication with user

8
A Typology of Expert System Tasks

• Interpretation Infer situation descriptions and
  meaning from sensed inputs
• Prediction Forecasting likely consequences of
  actions given certain input values
• Diagnosis Inferring faults/diseases based on
  interpretation of data
• Planning Suggest actions and plans to achieve
  stated goals
• Design Configuring object specifications for
  satisfying input requirements
• Prescription Prescribing remedies and treatments
  for malfunctions/diseases
• Monitoring Comparing sensory/user inputs to
  expected outcomes
• Control Governance of overall system behavior
• Instruction Diagnosis, prescription, and
  guidance of user behavior intelligent computer
  assisted learning

9
ES Inferencing (rule-based)

• Control cycle
• Match
• Select
• Execute rules
• Store facts
• Facts
• Known/deduced events or conditions, entered by
  user or deduced by system

• Rule examples
• Modos ponens
• If A then D
• Given A is true
• Deduce that D is true
• Modos tollens
• If B then (necessarily) C
• Given C is not true
• Deduce that B is not true

10
Inferencing Method

• Chaining--gtorganizing rules so that deduced fact
  from one is premise for the next
• Forward inductive, data-driven start with facts
  we know build toward goal
• Backward deductive, goal-driven, most commonly
  used in ES
• Choice of method based on
• Expert reasoning preference
• Efficiency
• Backward few goals or possible outcomes many
  known facts
• Forward many goals or possible outcomes few
  known facts

11
MYCIN

• 1970, Stanford, infectious agent identification
• Features
• Rule-based knowledge representation
• Probabilistic rules certainty factors
• Backward chaining inference logic
• Explanatory (questions results)
• User friendly interface

12
MYCIN example

• MYCIN What is the infection?
• USER Endarteritis
• MYCIN Is organism a rod or a coccus?
• USER Rod. . . .
• MYCIN Organism is E. coli
• USER How did you decide the organism is E.
  coli?
• MYCIN I used rule 084 to make conclusion. This
  gave a cumulative certainty factor of 0.51

13
Development of ES

• Task modeling
• Nature of problem, identify experts
• Performance modeling
• Knowledge acquisition and validation
• Interview experts/verbal protocal analysis/other
  techniques
• Research documentation
• Compare multiple sources
• Knowledge representation
• Knowledge engineer/analyst
• Decision trees/inference tables/rule lists
• Model uncertainty (rules probabilities)
• Prototyping

14
Validation Verification

• Knowledge base validation
• Marcot Table 8-5, pg. 254
• Characteristics of high quality knowledge base
• Responsibility of expert KE
• System verification
• Does the expert system accomplish its objectives?
• Systems analyst responsibility

15
Benefits

• Faster decisions
• Access to expert knowledge
• Increased productivity of experts
• Consistency
• Understanding
• Build organizational knowledge

16
Limitations

• Knowledge to build is not available, difficult to
  extract
• Logic may defy coding
• Humans are more flexible adaptive
• ES have limited sensory capabilities
• Human bias still a problem
• Social problems
• May be more error prone expensive to develop
  build